| 摘要 | 第1-7页 |
| ABSTRACT | 第7-12页 |
| CHAPTER 1 INTRODUCTION | 第12-26页 |
| ·RESEARCH PURPOSE AND SIGNIFICANCE | 第13-15页 |
| ·MOTION ANALYSIS AND BEHAVIOR RECOGNITION IN VIDEO DATA | 第15-18页 |
| ·Motion Analysis | 第15-17页 |
| ·Behavior Recognition | 第17-18页 |
| ·DESIGN ISSUES IN VISUAL ANALYSIS FOR BEHAVIOR MODELING | 第18-19页 |
| ·ABNORMAL BEHAVIOR DETECTION IN SURVEILLANCE SYSTEMS | 第19页 |
| ·CONTEXT-SPECIFIC CHALLENGES IN ABNORMAL BEHAVIOR DETECTION | 第19-22页 |
| ·Ambiguity of Definition | 第20-21页 |
| ·Noisy Data | 第21页 |
| ·Behavior Complexity | 第21-22页 |
| ·Limitation of Training Data | 第22页 |
| ·APPROACHES TO ABNORMAL BEHAVIOR DETECTION | 第22-24页 |
| ·THESIS ORGANIZATION | 第24-26页 |
| CHAPTER 2 BEHAVIOR REPRESENTATION | 第26-40页 |
| ·BEHAVIOR REPRESENTATION METHODS | 第26页 |
| ·FEATURES FOR BEHAVIOR REPRESENTATION | 第26-27页 |
| ·CROWD FLOW: AN APERIODIC DYNAMICAL SYSTEM | 第27-28页 |
| ·OPTICAL FLOW | 第28-30页 |
| ·POTENTIALS: INCOMPRESSIBLE AND IRROTATIONAL FLOW COMPONENTS | 第30-31页 |
| ·FINITE TIME LYAPUNOV EXPONENT (FTLE) | 第31-33页 |
| ·NOVEL DESCRIPTOR FOR MODELING CROWD DYNAMICS | 第33-34页 |
| ·FEATURE EXTRACTION FROM CROWD BEHAVIOR DATASET | 第34-38页 |
| ·SUMMARY | 第38-40页 |
| CHAPTER 3 BEHAVIOR PROFILING FROM BIO-INSPIRED CODEBOOKS | 第40-55页 |
| ·SEMANTIC REPRESENTATION: FROM FEATURES TO VISUAL CODEBOOK | 第40-41页 |
| ·Clustering and Clustering Algorithms | 第41-47页 |
| ·Conventional k-means Clustering | 第42-43页 |
| ·Nature Inspired Heuristics: Ants Clustering | 第43-47页 |
| ·Building a Visual Codebook Using Ant-Kmeans Co-clustering | 第47页 |
| ·EXPERIMENTS | 第47-53页 |
| ·Description of Datasets | 第47-53页 |
| ·Pre-processing and Feature Extraction | 第48-49页 |
| ·Feature Quantization and Codebook Formation | 第49-53页 |
| ·ANALYSIS OF RESULTS | 第53页 |
| ·Summary | 第53-55页 |
| CHAPTER 4 STATISTICAL MACHINE LEARNING FOR BEHAVIOR RECOGNITION | 第55-72页 |
| ·INTRODUCTION | 第55页 |
| ·BAYESIAN STATISTICAL MACHINE LEARNING | 第55-58页 |
| ·Modeling with the Exponential Family of Distributions | 第56-58页 |
| ·Maximum Likelihood Estimation | 第58页 |
| ·MAXIMUM A POSTERIORI ESTIMATE | 第58-59页 |
| ·THE DIRICHLET DISTRIBUTION | 第59-61页 |
| ·EXPECTATION MAXIMIZATION (EM) ESTIMATION FROM COUNTS | 第61-62页 |
| ·DISCOVERY OF BEHAVIOR PATTERNS USING TOPIC MODELS | 第62-65页 |
| ·Topic Decomposition and Document Generation using Video Data | 第63-65页 |
| ·LATENT DIRICHLET ALLOCATION (LDA) TOPIC MODEL | 第65-69页 |
| ·Model Parameters | 第66-68页 |
| ·Hyper-parameter and Posterior Distribution Estimation | 第68-69页 |
| ·APPLICATION TO CROWD DATASET | 第69-70页 |
| ·SUMMARY | 第70-72页 |
| CHAPTER 5 CONTEXTUAL ANOMALY DETECTION | 第72-98页 |
| ·ANOMALY DETECTION: LOCAL OBSERVATION TO GLOBAL INFERENCE | 第72-73页 |
| ·DESCRIPTION OF ABNORMAL BEHAVIOR DATASETS | 第73-74页 |
| ·EVALUATING THE BINARY DECISION PROBLEM | 第74-75页 |
| ·EXPERIMENTS FOR EARLY DETECTION OF ABNORMAL BEHAVIOR | 第75-93页 |
| ·Type 1 Abnormal Behavior. ‘RUSH’ | 第76-82页 |
| ·Classification of Frames for Test Clips Using Different Model Parameters | 第79-80页 |
| ·TYPE 1 Anomaly: Recall-Precision / ROC Analysis for Topic Models and Patch Sizes | 第80-82页 |
| ·Type 2 Abnormal Behaviors. ‘SCATTER’ | 第82-90页 |
| ·Classification of Frames for Test Clips using Different Model Parameters | 第85-87页 |
| ·TYPE 2 Anomalies: Recall-Precision / ROC Analysis for Topic Models and Patch Sizes | 第87-90页 |
| ·Type 3 Abnormal Behaviors. ‘HERDING’ | 第90-93页 |
| ·Classification of Frames for Test Clips using Different Topic Models | 第91-92页 |
| ·TYPE 3 Anomaly: Recall-Precision / ROC Analysis for Topic Models and Patch Sizes | 第92-93页 |
| ·EFFECT OF PATCH SIZE ON DETECTION ACCURACY FOR ANOMALY TYPES | 第93-96页 |
| ·Summary | 第96-98页 |
| Conclusion | 第98-100页 |
| References | 第100-111页 |
| Research Publications | 第111-112页 |
| Acknowledgements | 第112-113页 |
| Appendix | 第113-115页 |
